Y
Hacker News
new
|
ask
|
show
|
jobs
by
thesz
770 days ago
Note that KANs use LBFGS, which is second-order optimization method. My experience with the use of second-order features suggests that simple gradient descent often leads to divergence.